How SPUR wants to make AI companies report what they do with your journalism

Hi, I’m media innovation journalist Ulrike Langer and you’re reading my weekly AI in media newsletter. Thank you for being here! If someone has forwarded it to you please subscribe and never miss any future post.

My next European trip:
I’ll be in Europe again this fall and wanted to share a few dates in case it’s useful for you. I’m speaking at the Reporter:innen Workshop in Hamburg on September 25–26, at the Media Tech Hub Conference #MTHCON in Potsdam on September 29, and at the b°future Festival in Bonn on October 1–3. If you are at any of these events, please say hi! If you’re based in Europe and would like to set up a talk, workshop, or strategy session while I’m around, I still have some availability on October 5, 6, 7, and 8. If you’re interested please hit reply or send me a LinkedIn DM.

Over the past two issues, this newsletter has worked through two of the three questions that decide what AI is worth to a publisher. How TIME, The Economist and Dow Jones each decided what to expose to AI agents was about what you let the machines have. Why AI agents might finally be the perfect micro-payers was about how they might pay. This issue is about the question underneath both: How would you ever know what they took?

Right now, nobody on the selling side knows. The AI company can measure exactly which articles its systems fetch, how often, and for what purpose. The publisher signing the licensing deal sees none of that — and prices the deal anyway. That asymmetry is the problem the SPUR Coalition was founded to fix. SPUR stands for Standards for Publisher Usage Rights, and its answer is the Content Telemetry Standard. The standard's key architect knows the problem from the inside: David Buttle was heavily involved in the Financial Times' deal with OpenAI, one of the early agreements that, he says, were simply not built around usage data. His conclusion from that experience is now a draft standard whose public comment window closes this Friday, July 24 — which makes this the last week publishers can help shape it before version one hardens. I talked to Buttle on Tuesday.

A publisher's own visibility ends at the front door. Server logs record that a request happened — nothing about what happened next, and not even every request, because AI systems lean heavily on offline caches. A model can serve your reporting a thousand times from a copy it fetched once, and your logs show a single hit from months ago.

The  telemetry standard replaces that blindness with five reportable events that follow content through an AI system. 

The purpose of these five distinct events is not surveillance but pricing. Together they produce the metadata a publisher needs to make commercial decisions — whether to license at all, on what terms, at what price. Every deal on the books today was signed without that data. The standard exists so the data will be available for the next generation of deals.

The standard's weakness is stated in its own text: Four of the five events have no independent observer. The draft's one corroboration mechanism covers retrieval only — a publisher's own infrastructure can confirm that a fetch happened. Everything downstream — grounding, citation, display, engagement — is reported by the party that would be billed for it. The spec goes further still: Even cryptographic signing, once required, would prove who reported an event, not that the event is true or that all qualifying events were reported.

Buttle doesn't dispute the problem. His answer starts with contracts and middlemen: Negotiate audit rights into licensing deals, and let an intermediary layer between supply and demand do the checking on publishers' behalf. Behind that sits a market argument — the system must develop so that accurate data is advantageous to both sides. The argument also has a precedent - the ad market, which went from rampant fraud to a far smaller problem as verification became infrastructure. "We're on that journey with licensing," he says.

What he pointedly won't claim is that the system corrects itself. SPUR is not presupposing what enforcement looks like; it publishes a standard and expects an ecosystem of analytics, decisioning and audit intermediaries to form around it. The division of labor is deliberate: The standard measures. Someone else, eventually, verifies. Until that someone exists, telemetry numbers are directional intelligence, not invoice-grade truth.

The strongest argument against all of this comes from a publisher who already got paid. Le Monde's chief executive Louis Dreyfus signed guaranteed-minimum deals with OpenAI, Perplexity and Meta. At the WAN-IFRA congress in Marseille in early June, he asked why his company should join a coalition just to "be on the passenger seat and pay fees.” 

Buttle shared a stage with Dreyfus in Marseille earlier this year and had this debate live. His answer comes in three parts. He reclassifies the deal: What Le Monde signed is "basically a risk mitigation deal from the other side" — OpenAI paying off a publisher who might otherwise present a legal threat. He removes the comfort: "They will use his content regardless of whether he signs that deal. It gives the illusion of control when you don't really have control." And he dissolves the either/or: Take the money on the table, by all means — the proof that direct deals and collective rails coexist sits inside SPUR's own membership. The Guardian has an OpenAI deal. The FT has an OpenAI deal. 

And on July 9 the Associated Press — a company whose core business is licensing — joined as a founding member, adding a U.S. name to a roster of BBC, FT, Guardian, Sky, Telegraph, Mediahuis, Bonnier News and about 30 other members. The industry's most experienced seller of content licenses looked at collective usage rails and decided it needed them. That's a strong endorsement of SPUR. 

When does any of this produce revenue? Cosmin Ene, whose pay-per-usage case ran in last week's issue, expects nothing meaningful for 12 to 24 months. Buttle offers no forecast at all — too much depends on how the market forms. What he identifies instead is the event that would compress every timeline: a precedent-setting US court ruling on retrieval-augmented generation. The moment RAG infringement carries real legal cost, licensing stops being optional economics and becomes the cheaper path — "that's likely to force a licensing market into existence."

Publishers can't schedule a court ruling. What they can influence is mass: The more of the industry speaks with a single voice, the faster the change, and SPUR takes journalistic organizations of any size. (RSL — the rights-declaration leg of this stack, and Ene's current cause — gets its own issue here soon.)

The list of committed implementers is real but lopsided. Licensing and infrastructure intermediaries TollBit, Redpine and Monetization OS have told the coalition they plan to build the standard into their products; Microsoft and Fastly attended SPUR's London feedback workshop. Missing is the party the standard is written to measure. All five events are generated inside the AI companies' own systems and sent out from there — and no frontier lab has committed to sending them.

Buttle calls the sequencing deliberate: Start with the players connecting the two sides of the market, then move to the AI companies themselves, where he reports "super interesting conversations with the largest players." His case for why the labs will come is half moral, half commercial. The moral argument: Publishers employ journalists at great expense and have the right to know who uses the work, and to decide where and at what price. The commercial argument: The market will develop so that an AI company unwilling to transmit usage data is "unlikely to be able to access the most valuable IP on the internet,” he told me. Telemetry, in his telling, becomes table stakes: the ad market counts and prices in impressions, and the licensing market will count and price in usage events.

Whether the labs accept that unit is the whole game. Buttle wouldn't say which of them is at the table. What the standard's design does reveal is how SPUR intends to watch them either way.